SoftBank committed up to seventy-five billion euros this week to build five gigawatts of AI data centers in France — its largest such commitment in Europe — and the reason is not French talent, French capital, or French law. It is French electricity. France draws roughly seventy percent of its power from nuclear reactors, exports more electricity than any nation on Earth, and sells industrial power at less than half the price of Britain. The state-owned nuclear utility is partnering on the buildout and handing SoftBank the site of a former power plant to build upon. The most advanced computing infrastructure of the age is being sited not where the engineers are, nor where the customers are, but where the atoms are split — and the map of where artificial intelligence gets built is being redrawn, in real time, around the single resource the buildout cannot manufacture: power.
The Geography of the Watt
The constraint that drove Meta to reserve a gigawatt of solar power from orbit is now redrawing the map of the world, and it is doing so in favor of an unfashionable past. France built its nuclear fleet in the nineteen-seventies and eighties, in the aftermath of the oil shocks, a decision criticized for decades as the rest of the wealthy world turned against nuclear power on grounds of cost and fear. That fleet is the reason a Japanese conglomerate is now putting the equivalent of eighty-seven billion dollars into French soil. The first phase alone will deliver more than three gigawatts in the north of the country by 2031, built in part on the grounds of a decommissioned power station that the state utility is providing for the purpose. The energy decision of fifty years ago has become the compute decision of this one.
This is the inversion of the trend that defined the last half-century of industry. For decades, manufacturing fled the high-cost West for the cheap labor of Asia; the binding constraint was the price of human hands, and production migrated to wherever the hands were cheapest. The binding constraint has now changed. The most valuable manufacturing of the present age — the production of intelligence — is not labor-intensive but power-intensive, and so it migrates not toward cheap labor but toward cheap and abundant electricity. The factories of the mind go where the watts are, and the watts are in France, because France made, decades ago and for entirely unrelated reasons, the unglamorous decision to build more power than it needed.
The contrast with the United States is the part that should sting Washington. In America the buildout collides with a straining grid and communities fighting their power bills, the ratepayers charged for infrastructure they did not request, the jurisdictions blocking the data centers, the company reaching past the atmosphere because the ground cannot supply the load. In France the state nuclear utility hands the developer a former plant site and the president hosts the announcement at a summit designed to attract exactly this investment. Two energy futures, two welcomes. The American grid says no, slowly, in the form of a bill and a zoning fight; the French grid says yes, immediately, in the form of a decommissioned reactor’s footprint and a price under half the neighbors’.
The Sovereignty of the Grid
The location of compute is becoming a question of energy sovereignty, and energy sovereignty is decided on a timescale that mocks the quarterly horizon of the industry it now governs. A nuclear plant takes more than a decade to design, license, and build. The electricity the buildout needs is needed now. These two facts cannot be reconciled by any amount of capital, which means the nations that will host the most valuable infrastructure of the century are, to a large degree, already determined — fixed by power decisions made years or decades ago, by planners who could not have imagined a data center and were solving for entirely different problems. The opportunity of the future is being allocated by the energy policy of the past, and the allocation is largely already complete.
This is a brutal lesson for the countries that deferred the hard choice. Building abundant power is expensive, slow, politically painful, and unrewarded for years — the kind of decision a democracy on a short electoral cycle is structurally bad at making, because the costs fall on the present and the benefits accrue to a future no incumbent will be in office to claim. France made that decision under conditions that made it possible, and bore the criticism for fifty years, and is now collecting a return its original planners never imagined. The nations that chose the cheaper, easier path — that let nuclear stagnate, that built no surplus, that assumed power would always be available when needed — are discovering that they deferred not just the power but their place in the industry the power now decides.
And the resource cannot be conjured on demand. This is what separates the energy constraint from the others the buildout has hit. A chip shortage is solved in a couple of years; a fiber shortage in two; even a workforce can be trained inside a decade. Abundant baseload power is the work of a generation, and a nation that does not have it when the demand arrives cannot acquire it in time to matter for this wave. The compute will go where the power already is, and stay there, and the economic gravity will follow, and the countries watching the buildout route around their grids to France’s reactors are watching, in real time, the price of an unglamorous decision they declined to make when it was theirs to make.
What This Means
There is a clarifying humility in this, of the kind the buildout rarely permits. Artificial intelligence is the most modern thing the species builds, spoken of in the language of transcendence and the singularity and the reshaping of every institution — and it has just announced that where it physically lives is determined by the most ancient and least glamorous decision a society makes, which is whether to generate more power than it consumes. The frontier of the mind turns out to be sited by the infrastructure of the grid, and the grid is poured in concrete over decades by engineers no one celebrates, solving for a future they will not see.
The deeper point is about which decisions actually matter, and on what horizon. The industry conducts itself as though everything important happens at the speed of a model release, a funding round, a quarterly print. But the thing now determining the geography of the entire enterprise was decided fifty years ago, slowly, against opposition, by a state willing to spend the present for the future — and nothing the present can do will change it in time. The reactors that matter for this wave are already built or already absent, and a nation discovers which it is not by debating energy policy now but by watching where the eighty-seven-billion-dollar commitments land.
I observed of the fiber shortage that the future is gated by the furnace. It is gated, more fundamentally, by the reactor, and the reactors that govern this decade were poured in concrete in the last century by planners the present has mostly forgotten. France’s bet on the atom was made by people solving for oil shocks and national independence, with no notion of what a data center was or that the most valuable industry of their grandchildren’s era would one day come looking for exactly the surplus they were building. They could not have known. They built it anyway, and the building of it is now quietly deciding which nations get to host the future, on a timescale that no quarterly report and no funding round can touch. The most futuristic technology in the world is shopping for the oldest thing in the economy, and it is finding it in the places that chose, long ago and unfashionably, to make more of it than they needed.